libmango: types: add macro to define a kern_msg_handle_t

include/kernel/arg.h

@@ -1,8 +1,8 @@
 #ifndef KERNEL_ARG_H_
 #define KERNEL_ARG_H_
 
+#include <mango/types.h>
 #include <stdbool.h>
-#include <mango/status.h>
 
 #define CMDLINE_MAX 4096
 

include/kernel/channel.h

@@ -25,15 +25,12 @@ extern kern_status_t channel_enqueue_msg(
 
 extern kern_status_t channel_recv_msg(
 	struct channel *channel,
-	msgid_t *out_id,
-	struct iovec *out_data,
-	size_t out_data_count,
+	kern_msg_t *out_msg,
 	unsigned long *irq_flags);
 extern kern_status_t channel_reply_msg(
 	struct channel *channel,
 	msgid_t id,
-	const struct iovec *resp_data,
-	size_t resp_data_count,
+	const kern_msg_t *reply,
 	unsigned long *irq_flags);
 
 extern kern_status_t channel_read_msg(
@@ -41,7 +38,7 @@ extern kern_status_t channel_read_msg(
 	msgid_t msg,
 	size_t offset,
 	struct vm_region *dest_region,
-	const struct iovec *dest_iov,
+	const kern_iovec_t *dest_iov,
 	size_t dest_iov_count,
 	size_t *nr_read);
 extern kern_status_t channel_write_msg(
@@ -49,7 +46,7 @@ extern kern_status_t channel_write_msg(
 	msgid_t msg,
 	size_t offset,
 	struct vm_region *src_region,
-	const struct iovec *src_iov,
+	const kern_iovec_t *src_iov,
 	size_t src_iov_count,
 	size_t *nr_written);
 

include/kernel/console.h

@@ -14,9 +14,10 @@
    representing a serial port may allow both sending AND receiving over the
    port.
 */
-#include <kernel/queue.h>
 #include <kernel/locks.h>
+#include <kernel/queue.h>
 #include <mango/status.h>
+#include <mango/types.h>
 
 #ifdef __cplusplus
 extern "C" {

include/kernel/handle.h

@@ -3,6 +3,7 @@
 
 #include <kernel/bitmap.h>
 #include <mango/status.h>
+#include <mango/types.h>
 #include <stddef.h>
 #include <stdint.h>
 
@@ -16,6 +17,7 @@ typedef uintptr_t handle_flags_t;
 
 struct task;
 struct object;
+struct vm_region;
 struct handle_list;
 
 struct handle {
@@ -54,4 +56,14 @@ extern struct handle *handle_table_get_handle(
 	struct handle_table *tab,
 	kern_handle_t handle);
 
+extern kern_status_t handle_table_transfer(
+	struct vm_region *dst_region,
+	struct handle_table *dst,
+	kern_msg_handle_t *dst_handles,
+	size_t dst_handles_max,
+	struct vm_region *src_region,
+	struct handle_table *src,
+	kern_msg_handle_t *src_handles,
+	size_t src_handles_count);
+
 #endif

include/kernel/iovec.h

@@ -8,7 +8,7 @@ struct iovec_iterator {
 	/* if this is set, we are iterating over a list of iovecs stored in
 	 * userspace, and must go through this region to retrieve the data. */
 	struct vm_region *it_region;
-	const struct iovec *it_vecs;
+	const kern_iovec_t *it_vecs;
 	size_t it_nr_vecs;
 	size_t it_vec_ptr;
 
@@ -18,12 +18,12 @@ struct iovec_iterator {
 
 extern void iovec_iterator_begin(
 	struct iovec_iterator *it,
-	const struct iovec *vecs,
+	const kern_iovec_t *vecs,
 	size_t nr_vecs);
 extern void iovec_iterator_begin_user(
 	struct iovec_iterator *it,
 	struct vm_region *address_space,
-	const struct iovec *vecs,
+	const kern_iovec_t *vecs,
 	size_t nr_vecs);
 
 extern void iovec_iterator_seek(struct iovec_iterator *it, size_t nr_bytes);

include/kernel/locks.h

@@ -22,6 +22,38 @@ typedef __aligned(8) ml_hwlock_t spin_lock_t;
 #define spin_unlock_irqrestore(lck, flags)                                     \
 	ml_hwlock_unlock_irqrestore(lck, flags);
 
+static inline void spin_lock_pair_irqsave(
+	spin_lock_t *a,
+	spin_lock_t *b,
+	unsigned long *flags)
+{
+	if (a == b) {
+		spin_lock_irqsave(a, flags);
+	} else if (a < b) {
+		spin_lock_irqsave(a, flags);
+		spin_lock(b);
+	} else {
+		spin_lock_irqsave(b, flags);
+		spin_lock(a);
+	}
+}
+
+static inline void spin_unlock_pair_irqrestore(
+	spin_lock_t *a,
+	spin_lock_t *b,
+	unsigned long flags)
+{
+	if (a == b) {
+		spin_unlock_irqrestore(a, flags);
+	} else if (a < b) {
+		spin_unlock(b);
+		spin_unlock_irqrestore(a, flags);
+	} else {
+		spin_unlock(a);
+		spin_unlock_irqrestore(b, flags);
+	}
+}
+
 #ifdef __cplusplus
 }
 #endif

include/kernel/msg.h

@@ -23,10 +23,7 @@ struct msg {
 	kern_status_t msg_result;
 	struct port *msg_sender_port;
 	struct thread *msg_sender_thread;
-	const struct iovec *msg_req_data;
-	size_t msg_req_data_count;
-	struct iovec *msg_resp_data;
-	size_t msg_resp_data_count;
+	kern_msg_t msg_req, msg_resp;
 };
 
 #endif

include/kernel/port.h

@@ -32,10 +32,8 @@ extern kern_status_t port_connect(struct port *port, struct channel *remote);
 extern kern_status_t port_disconnect(struct port *port);
 extern kern_status_t port_send_msg(
 	struct port *port,
-	const struct iovec *req_data,
-	size_t req_data_count,
-	struct iovec *resp_data,
-	size_t resp_data_count,
+	const kern_msg_t *msg,
+	kern_msg_t *out_response,
 	unsigned long *lock_flags);
 
 DEFINE_OBJECT_LOCK_FUNCTION(port, p_base)

include/kernel/sched.h

@@ -85,6 +85,7 @@ struct task {
 
 	pmap_t t_pmap;
 	struct vm_region *t_address_space;
+	spin_lock_t t_handles_lock;
 	struct handle_table *t_handles;
 	struct btree b_channels;
 

include/kernel/syscall.h

@@ -153,32 +153,26 @@ extern kern_status_t sys_port_disconnect(kern_handle_t port);
 
 extern kern_status_t sys_msg_send(
 	kern_handle_t port,
-	const struct iovec *req_data,
-	size_t req_data_count,
-	struct iovec *resp_data,
-	size_t resp_data_count);
-extern kern_status_t sys_msg_recv(
-	kern_handle_t channel,
-	msgid_t *out_id,
-	struct iovec *out_data,
-	size_t out_data_count);
+	const kern_msg_t *msg,
+	kern_msg_t *out_reply);
+extern kern_status_t sys_msg_recv(kern_handle_t channel, kern_msg_t *out_msg);
+
 extern kern_status_t sys_msg_reply(
 	kern_handle_t channel,
 	msgid_t id,
-	const struct iovec *reply_data,
-	size_t reply_data_count);
+	const kern_msg_t *msg);
 extern kern_status_t sys_msg_read(
 	kern_handle_t channel_handle,
 	msgid_t id,
 	size_t offset,
-	const struct iovec *iov,
+	const kern_iovec_t *iov,
 	size_t iov_count,
 	size_t *nr_read);
 extern kern_status_t sys_msg_write(
 	kern_handle_t channel,
 	msgid_t id,
 	size_t offset,
-	const struct iovec *in,
+	const kern_iovec_t *in,
 	size_t nr_in,
 	size_t *nr_written);
 

include/kernel/vm-region.h

@@ -157,6 +157,15 @@ extern kern_status_t vm_region_read_kernel(
 	void *dest,
 	size_t *nr_read);
 
+/* write data to the user-space area of a vm-region from a kernel-mode buffer
+ */
+extern kern_status_t vm_region_write_kernel(
+	struct vm_region *dst_region,
+	virt_addr_t dst_ptr,
+	size_t count,
+	const void *src,
+	size_t *nr_written);
+
 extern kern_status_t vm_region_memmove(
 	struct vm_region *dest_region,
 	virt_addr_t dest_ptr,
@@ -168,11 +177,11 @@ extern kern_status_t vm_region_memmove(
 extern kern_status_t vm_region_memmove_v(
 	struct vm_region *dest_region,
 	size_t dest_offset,
-	const struct iovec *dest,
+	const kern_iovec_t *dest,
 	size_t nr_dest,
 	struct vm_region *src_region,
 	size_t src_offset,
-	const struct iovec *src,
+	const kern_iovec_t *src,
 	size_t nr_src,
 	size_t bytes_to_move,
 	size_t *nr_bytes_moved);

kernel/arg.c

@@ -1,6 +1,7 @@
 #include <kernel/arg.h>
-#include <kernel/libc/string.h>
 #include <kernel/libc/ctype.h>
+#include <kernel/libc/string.h>
+#include <mango/status.h>
 
 static char g_cmdline[CMDLINE_MAX + 1] = {0};
 
@@ -81,7 +82,6 @@ static char *advance_to_next_arg(char *s, char *max)
 	return s;
 }
 
-
 const char *arg_value(const char *arg_name)
 {
 	char *s = g_cmdline;

kernel/bsp.c

@@ -69,7 +69,7 @@ kern_status_t bsp_load(struct bsp *bsp, const struct boot_module *mod)
 	return KERN_OK;
 }
 
-static kern_status_t map_executable(
+static kern_status_t map_executable_dyn(
 	struct bsp *bsp,
 	struct task *task,
 	virt_addr_t *entry)
@@ -164,6 +164,82 @@ static kern_status_t map_executable(
 	return KERN_OK;
 }
 
+static kern_status_t map_executable_exec(
+	struct bsp *bsp,
+	struct task *task,
+	virt_addr_t *entry)
+{
+	kern_status_t status = KERN_OK;
+	struct vm_object *data = vm_object_create(
+		".data",
+		5,
+		bsp->bsp_trailer.bsp_data_size,
+		VM_PROT_READ | VM_PROT_WRITE | VM_PROT_USER);
+	if (!data) {
+		return KERN_NO_MEMORY;
+	}
+
+	virt_addr_t text_base = 0, data_base = 0;
+
+	off_t text_foffset = bsp->bsp_trailer.bsp_exec_offset
+	                   + bsp->bsp_trailer.bsp_text_faddr;
+	off_t data_foffset = 0;
+	off_t text_voffset = bsp->bsp_trailer.bsp_text_vaddr;
+	off_t data_voffset = bsp->bsp_trailer.bsp_data_vaddr;
+
+	text_voffset -= vm_region_get_base_address(task->t_address_space);
+	data_voffset -= vm_region_get_base_address(task->t_address_space);
+
+#if 0
+	size_t tmp = 0;
+	status = vm_object_copy(
+		data,
+		0,
+		bsp->bsp_vmo,
+		bsp->bsp_trailer.bsp_data_faddr,
+		bsp->bsp_trailer.bsp_data_size,
+		&tmp);
+
+	tracek("read %zuB of data from executable", tmp);
+#endif
+
+	tracek("text_foffset=%06llx, data_foffset=%06llx",
+	       text_foffset,
+	       data_foffset);
+	tracek("text_voffset=%08llx, data_voffset=%08llx",
+	       text_voffset,
+	       data_voffset);
+
+	status = vm_region_map_object(
+		task->t_address_space,
+		text_voffset,
+		bsp->bsp_vmo,
+		text_foffset,
+		bsp->bsp_trailer.bsp_text_size,
+		VM_PROT_READ | VM_PROT_EXEC | VM_PROT_USER,
+		&text_base);
+	if (status != KERN_OK) {
+		return status;
+	}
+
+	status = vm_region_map_object(
+		task->t_address_space,
+		data_voffset,
+		data,
+		data_foffset,
+		bsp->bsp_trailer.bsp_data_size,
+		VM_PROT_READ | VM_PROT_WRITE | VM_PROT_USER,
+		&data_base);
+	if (status != KERN_OK) {
+		return status;
+	}
+
+	tracek("text_base=%08llx, data_base=%08llx", text_base, data_base);
+
+	*entry = bsp->bsp_trailer.bsp_exec_entry;
+	return KERN_OK;
+}
+
 kern_status_t bsp_launch_async(struct bsp *bsp, struct task *task)
 {
 	virt_addr_t stack_buffer, bsp_data_base;
@@ -205,7 +281,7 @@ kern_status_t bsp_launch_async(struct bsp *bsp, struct task *task)
 		return status;
 	}
 
-	status = map_executable(bsp, task, &entry);
+	status = map_executable_exec(bsp, task, &entry);
 	if (status != KERN_OK) {
 		return status;
 	}

kernel/channel.c

@@ -124,9 +124,7 @@ extern kern_status_t channel_enqueue_msg(
 
 extern kern_status_t channel_recv_msg(
 	struct channel *channel,
-	msgid_t *out_id,
-	struct iovec *out_data,
-	size_t out_data_count,
+	kern_msg_t *out_msg,
 	unsigned long *irq_flags)
 {
 	struct wait_item waiter;
@@ -155,20 +153,47 @@ extern kern_status_t channel_recv_msg(
 
 	struct vm_region *src = sender->t_address_space,
 			 *dst = receiver->t_address_space;
+
 	unsigned long f;
 	vm_region_lock_pair_irqsave(src, dst, &f);
 
 	kern_status_t status = vm_region_memmove_v(
 		dst,
 		0,
-		out_data,
-		out_data_count,
+		out_msg->msg_data,
+		out_msg->msg_data_count,
 		src,
 		0,
-		msg->msg_req_data,
-		msg->msg_req_data_count,
+		msg->msg_req.msg_data,
+		msg->msg_req.msg_data_count,
 		VM_REGION_COPY_ALL,
 		NULL);
+
+	if (status != KERN_OK) {
+		kmsg_reply_error(msg, status, &msg_lock_flags);
+		return status;
+	}
+
+	struct handle_table *src_table = sender->t_handles,
+			    *dst_table = receiver->t_handles;
+
+	spin_lock_pair_irqsave(
+		&sender->t_handles_lock,
+		&receiver->t_handles_lock,
+		&f);
+	status = handle_table_transfer(
+		dst,
+		dst_table,
+		out_msg->msg_handles,
+		out_msg->msg_handles_count,
+		src,
+		src_table,
+		msg->msg_req.msg_handles,
+		msg->msg_req.msg_handles_count);
+	spin_unlock_pair_irqrestore(
+		&sender->t_handles_lock,
+		&receiver->t_handles_lock,
+		f);
 	vm_region_unlock_pair_irqrestore(src, dst, f);
 
 	if (status != KERN_OK) {
@@ -176,7 +201,9 @@ extern kern_status_t channel_recv_msg(
 		return status;
 	}
 
-	*out_id = msg->msg_id;
+	out_msg->msg_id = msg->msg_id;
+	out_msg->msg_sender = msg->msg_sender_thread->tr_parent->t_id;
+	out_msg->msg_endpoint = msg->msg_sender_port->p_base.ob_id;
 
 	spin_unlock_irqrestore(&msg->msg_lock, msg_lock_flags);
 
@@ -186,8 +213,7 @@ extern kern_status_t channel_recv_msg(
 extern kern_status_t channel_reply_msg(
 	struct channel *channel,
 	msgid_t id,
-	const struct iovec *resp_data,
-	size_t resp_data_count,
+	const kern_msg_t *reply,
 	unsigned long *irq_flags)
 {
 	unsigned long msg_lock_flags;
@@ -216,14 +242,40 @@ extern kern_status_t channel_reply_msg(
 	kern_status_t status = vm_region_memmove_v(
 		dst,
 		0,
-		msg->msg_resp_data,
-		msg->msg_resp_data_count,
+		msg->msg_resp.msg_data,
+		msg->msg_resp.msg_data_count,
 		src,
 		0,
-		resp_data,
-		resp_data_count,
+		reply->msg_data,
+		reply->msg_data_count,
 		VM_REGION_COPY_ALL,
 		NULL);
+
+	if (status != KERN_OK) {
+		kmsg_reply_error(msg, status, &msg_lock_flags);
+		return status;
+	}
+
+	struct handle_table *src_table = sender->t_handles,
+			    *dst_table = receiver->t_handles;
+
+	spin_lock_pair_irqsave(
+		&sender->t_handles_lock,
+		&receiver->t_handles_lock,
+		&f);
+	status = handle_table_transfer(
+		dst,
+		dst_table,
+		msg->msg_resp.msg_handles,
+		msg->msg_resp.msg_handles_count,
+		src,
+		src_table,
+		reply->msg_handles,
+		reply->msg_handles_count);
+	spin_unlock_pair_irqrestore(
+		&sender->t_handles_lock,
+		&receiver->t_handles_lock,
+		f);
 	vm_region_unlock_pair_irqrestore(src, dst, f);
 
 	if (status != KERN_OK) {
@@ -241,7 +293,7 @@ extern kern_status_t channel_read_msg(
 	msgid_t id,
 	size_t offset,
 	struct vm_region *dest_region,
-	const struct iovec *dest_iov,
+	const kern_iovec_t *dest_iov,
 	size_t dest_iov_count,
 	size_t *nr_read)
 {
@@ -270,8 +322,8 @@ extern kern_status_t channel_read_msg(
 		dest_iov_count,
 		src_region,
 		offset,
-		msg->msg_req_data,
-		msg->msg_req_data_count,
+		msg->msg_req.msg_data,
+		msg->msg_req.msg_data_count,
 		VM_REGION_COPY_ALL,
 		nr_read);
 	vm_region_unlock_pair_irqrestore(src_region, dest_region, f);
@@ -286,7 +338,7 @@ extern kern_status_t channel_write_msg(
 	msgid_t id,
 	size_t offset,
 	struct vm_region *src_region,
-	const struct iovec *src_iov,
+	const kern_iovec_t *src_iov,
 	size_t src_iov_count,
 	size_t *nr_written)
 {
@@ -310,11 +362,11 @@ extern kern_status_t channel_write_msg(
 
 	kern_status_t status = vm_region_memmove_v(
 		dest_region,
-		0,
-		msg->msg_resp_data,
-		msg->msg_resp_data_count,
-		src_region,
 		offset,
+		msg->msg_resp.msg_data,
+		msg->msg_resp.msg_data_count,
+		src_region,
+		0,
 		src_iov,
 		src_iov_count,
 		VM_REGION_COPY_ALL,

kernel/handle.c

@@ -3,7 +3,9 @@
 #include <kernel/object.h>
 #include <kernel/sched.h>
 #include <kernel/util.h>
+#include <kernel/vm-region.h>
 #include <kernel/vm.h>
+#include <mango/types.h>
 
 /* depth=3 gives a maximum of ~66.6 million handles */
 #define MAX_TABLE_DEPTH  3
@@ -191,3 +193,144 @@ struct handle *handle_table_get_handle(
 
 	return &tab->t_handles.t_handle_list[handle_index];
 }
+
+kern_status_t handle_table_transfer(
+	struct vm_region *dst_region,
+	struct handle_table *dst,
+	kern_msg_handle_t *dst_handles,
+	size_t dst_handles_max,
+	struct vm_region *src_region,
+	struct handle_table *src,
+	kern_msg_handle_t *src_handles,
+	size_t src_handles_count)
+{
+	kern_status_t status = KERN_OK;
+	size_t to_transfer = MIN(dst_handles_max, src_handles_count);
+
+	size_t i = 0;
+	for (size_t i = 0; i < to_transfer; i++) {
+		kern_msg_handle_t src_handle = {0}, dst_handle = {0};
+		virt_addr_t src_handle_addr
+			= (virt_addr_t)src_handles + (i * sizeof src_handle);
+		virt_addr_t dst_handle_addr
+			= (virt_addr_t)dst_handles + (i * sizeof dst_handle);
+		status = vm_region_read_kernel(
+			src_region,
+			src_handle_addr,
+			sizeof src_handle,
+			&src_handle,
+			NULL);
+
+		if (status != KERN_OK) {
+			src_handle.hnd_result = KERN_OK;
+			vm_region_write_kernel(
+				src_region,
+				src_handle_addr,
+				sizeof src_handle,
+				&src_handle,
+				NULL);
+			break;
+		}
+
+		struct handle *src_entry
+			= handle_table_get_handle(src, src_handle.hnd_value);
+		struct handle *dst_entry = NULL;
+		kern_handle_t dst_value = KERN_HANDLE_INVALID;
+
+		if (!src_entry) {
+			status = KERN_INVALID_ARGUMENT;
+			src_handle.hnd_result = KERN_OK;
+			vm_region_write_kernel(
+				src_region,
+				src_handle_addr,
+				sizeof src_handle,
+				&src_handle,
+				NULL);
+			break;
+		}
+
+		switch (src_handle.hnd_mode) {
+		case KERN_MSG_HANDLE_IGNORE:
+			break;
+		case KERN_MSG_HANDLE_MOVE:
+			status = handle_table_alloc_handle(
+				dst,
+				&dst_entry,
+				&dst_value);
+			if (status != KERN_OK) {
+				break;
+			}
+
+			dst_entry->h_object = src_entry->h_object;
+			dst_entry->h_flags = src_entry->h_flags;
+			object_add_handle(dst_entry->h_object);
+
+			handle_table_free_handle(src, src_handles[i].hnd_value);
+
+			dst_handle.hnd_mode = src_handles[i].hnd_mode;
+			dst_handle.hnd_value = dst_value;
+			dst_handle.hnd_result = KERN_OK;
+			break;
+		case KERN_MSG_HANDLE_COPY:
+			status = handle_table_alloc_handle(
+				dst,
+				&dst_entry,
+				&dst_value);
+			if (status != KERN_OK) {
+				break;
+			}
+
+			dst_entry->h_object = src_entry->h_object;
+			dst_entry->h_flags = src_entry->h_flags;
+			object_add_handle(dst_entry->h_object);
+
+			dst_handle.hnd_mode = src_handles[i].hnd_mode;
+			dst_handle.hnd_value = dst_value;
+			dst_handle.hnd_result = KERN_OK;
+			break;
+		default:
+			status = KERN_INVALID_ARGUMENT;
+			break;
+		}
+
+		src_handle.hnd_result = status;
+
+		vm_region_write_kernel(
+			src_region,
+			src_handle_addr,
+			sizeof src_handle,
+			&src_handle,
+			NULL);
+		vm_region_write_kernel(
+			dst_region,
+			dst_handle_addr,
+			sizeof dst_handle,
+			&dst_handle,
+			NULL);
+	}
+
+	for (; i < src_handles_count; i++) {
+		kern_msg_handle_t handle = {0};
+		virt_addr_t handle_addr
+			= (virt_addr_t)src_handles + (i * sizeof handle);
+		vm_region_read_kernel(
+			src_region,
+			handle_addr,
+			sizeof handle,
+			&handle,
+			NULL);
+
+		if (handle.hnd_mode != KERN_MSG_HANDLE_MOVE) {
+			continue;
+		}
+
+		struct handle *src_entry
+			= handle_table_get_handle(src, handle.hnd_value);
+		if (src_entry) {
+			object_remove_handle(src_entry->h_object);
+			handle_table_free_handle(src, handle.hnd_value);
+		}
+	}
+
+	return status;
+}

kernel/iovec.c

@@ -6,7 +6,7 @@
 static bool read_iovec(
 	struct iovec_iterator *it,
 	size_t index,
-	struct iovec *out)
+	kern_iovec_t *out)
 {
 	if (index >= it->it_nr_vecs) {
 		return false;
@@ -20,18 +20,18 @@ static bool read_iovec(
 	size_t nr_read = 0;
 	kern_status_t status = vm_region_read_kernel(
 		it->it_region,
-		(virt_addr_t)it->it_vecs + (index * sizeof(struct iovec)),
-		sizeof(struct iovec),
+		(virt_addr_t)it->it_vecs + (index * sizeof(kern_iovec_t)),
+		sizeof(kern_iovec_t),
 		out,
 		&nr_read);
 
-	return (status == KERN_OK && nr_read != sizeof(struct iovec));
+	return (status == KERN_OK && nr_read != sizeof(kern_iovec_t));
 }
 
 void iovec_iterator_begin_user(
 	struct iovec_iterator *it,
 	struct vm_region *region,
-	const struct iovec *vecs,
+	const kern_iovec_t *vecs,
 	size_t nr_vecs)
 {
 	memset(it, 0x0, sizeof *it);
@@ -39,7 +39,7 @@ void iovec_iterator_begin_user(
 	it->it_vecs = vecs;
 	it->it_nr_vecs = nr_vecs;
 
-	struct iovec iov;
+	kern_iovec_t iov;
 
 	while (it->it_vec_ptr < nr_vecs) {
 		read_iovec(it, it->it_vec_ptr, &iov);
@@ -60,7 +60,7 @@ void iovec_iterator_begin_user(
 
 void iovec_iterator_begin(
 	struct iovec_iterator *it,
-	const struct iovec *vecs,
+	const kern_iovec_t *vecs,
 	size_t nr_vecs)
 {
 	memset(it, 0x0, sizeof *it);
@@ -97,7 +97,7 @@ void iovec_iterator_seek(struct iovec_iterator *it, size_t nr_bytes)
 		}
 
 		nr_bytes -= to_seek;
-		struct iovec iov;
+		kern_iovec_t iov;
 
 		it->it_vec_ptr++;
 		while (it->it_vec_ptr < it->it_nr_vecs) {

kernel/port.c

@@ -78,10 +78,8 @@ kern_status_t port_disconnect(struct port *port)
 
 kern_status_t port_send_msg(
 	struct port *port,
-	const struct iovec *req_data,
-	size_t req_data_count,
-	struct iovec *resp_data,
-	size_t resp_data_count,
+	const kern_msg_t *in_msg,
+	kern_msg_t *out_reply,
 	unsigned long *lock_flags)
 {
 	if (port->p_status != PORT_READY) {
@@ -94,10 +92,8 @@ kern_status_t port_send_msg(
 	msg->msg_status = KMSG_WAIT_RECEIVE;
 	msg->msg_sender_thread = self;
 	msg->msg_sender_port = port;
-	msg->msg_req_data = req_data;
-	msg->msg_req_data_count = req_data_count;
-	msg->msg_resp_data = resp_data;
-	msg->msg_resp_data_count = resp_data_count;
+	memcpy(&msg->msg_req, in_msg, sizeof msg->msg_req);
+	memcpy(&msg->msg_resp, out_reply, sizeof msg->msg_req);
 
 	unsigned long flags;
 	channel_lock_irqsave(port->p_remote, &flags);

kernel/status.c

@@ -1,4 +1,5 @@
 #include <mango/status.h>
+#include <mango/types.h>
 
 #define ERROR_STRING_CASE(code)                                                \
 	case code:                                                             \

libmango/include-user/mango/log.h

@@ -2,6 +2,7 @@
 #define MANGO_LOG_H_
 
 #include <mango/status.h>
+#include <mango/types.h>
 
 #undef TRACE
 
@@ -9,18 +10,18 @@ extern kern_status_t kern_log(const char *s);
 
 #define kern_logf(...)                                                         \
 	do {                                                                   \
-		char s[128];                                                   \
-		snprintf(s, sizeof s, __VA_ARGS__);                            \
-		kern_log(s);                                                   \
+		char __logbuf[128];                                            \
+		snprintf(__logbuf, sizeof __logbuf, __VA_ARGS__);              \
+		kern_log(__logbuf);                                            \
 	} while (0)
 
 #ifdef TRACE
 #define kern_trace(...) kern_log(__VA_ARGS__)
 #define kern_tracef(...)                                                       \
 	do {                                                                   \
-		char s[128];                                                   \
-		snprintf(s, sizeof s, __VA_ARGS__);                            \
-		kern_log(s);                                                   \
+		char __logbuf[128];                                            \
+		snprintf(__logbuf, sizeof __logbuf, __VA_ARGS__);              \
+		kern_log(__logbuf);                                            \
 	} while (0)
 #else
 #define kern_trace(...)

libmango/include-user/mango/msg.h

@@ -14,28 +14,21 @@ extern kern_status_t port_disconnect(kern_handle_t port);
 
 extern kern_status_t msg_send(
 	kern_handle_t port,
-	const struct iovec *req_data,
-	size_t req_data_count,
-	struct iovec *resp_data,
-	size_t resp_data_count);
+	const kern_msg_t *msg,
+	kern_msg_t *out_response);
 
-extern kern_status_t msg_recv(
-	kern_handle_t channel,
-	msgid_t *out_id,
-	struct iovec *out_data,
-	size_t out_data_count);
+extern kern_status_t msg_recv(kern_handle_t channel, kern_msg_t *out);
 
 extern kern_status_t msg_reply(
 	kern_handle_t channel,
 	msgid_t id,
-	const struct iovec *reply_data,
-	size_t reply_data_count);
+	const kern_msg_t *response);
 
 extern kern_status_t msg_read(
 	kern_handle_t channel,
 	msgid_t id,
 	size_t offset,
-	struct iovec *out,
+	kern_iovec_t *out,
 	size_t out_count,
 	size_t *nr_read);
 
@@ -43,7 +36,7 @@ extern kern_status_t msg_write(
 	kern_handle_t channel,
 	msgid_t id,
 	size_t offset,
-	const struct iovec *in,
+	const kern_iovec_t *in,
 	size_t nr_in,
 	size_t *nr_written);
 

libmango/include/mango/status.h

@@ -1,8 +1,6 @@
 #ifndef MANGO_STATUS_H_
 #define MANGO_STATUS_H_
 
-typedef unsigned int kern_status_t;
-
 #define KERN_OK               (0)
 #define KERN_UNIMPLEMENTED    (1)
 #define KERN_NAME_EXISTS      (2)

libmango/include/mango/types.h

@@ -12,27 +12,27 @@
 #define VM_PROT_NOCACHE        0x10u
 #define VM_PROT_MAP_SPECIFIC   0x40u
 
-/* if this flag is set, other tasks can connect to this channel using
- * the port_connect_* syscalls.
- * if this flag is NOT set, only threads in the task that owns the channel
- * can create ports connecting to it. */
-#define CHANNEL_F_ALLOW_DIRECT_CONNECTIONS 0x01u
-
-/* msg_reply: once the reply has been sent, disconnect the port that sent the
- * original message */
-#define MSG_F_DISCONNECT_AFTER_REPLY       0x01u
-
 #define VM_REGION_ANY_OFFSET   ((off_t) - 1)
 #define KERN_HANDLE_INVALID    ((kern_handle_t)0xFFFFFFFF)
 
 #define KERN_CFG_INVALID       0x00u
 #define KERN_CFG_PAGE_SIZE     0x01u
 
+#define KERN_MSG_MAX_HANDLES   64
+#define KERN_MSG_HANDLE_IGNORE 0
+#define KERN_MSG_HANDLE_MOVE   1
+#define KERN_MSG_HANDLE_COPY   2
+
 #define IOVEC(p, len)                                                          \
 	{                                                                      \
 		.io_base = (virt_addr_t)(p),                                   \
 		.io_len = (len),                                               \
 	}
+#define MSG_HANDLE(mode, value)                                                \
+	{                                                                      \
+		.hnd_mode = (mode),                                            \
+		.hnd_value = (value),                                          \
+	}
 #define MSG(data, data_count, handles, handles_len)                            \
 	{                                                                      \
 		.msg_data = (data),                                            \
@@ -47,15 +47,41 @@ typedef uint64_t msgid_t;
 typedef uint64_t off_t;
 typedef uint64_t koid_t;
 typedef unsigned int tid_t;
+typedef unsigned int kern_status_t;
 typedef uint32_t kern_handle_t;
 typedef uint32_t kern_config_key_t;
 typedef uint32_t vm_prot_t;
+typedef int64_t ssize_t;
 
 typedef unsigned int umode_t;
 
-struct iovec {
+typedef struct {
 	virt_addr_t io_base;
 	size_t io_len;
-};
+} kern_iovec_t;
+
+typedef struct {
+	unsigned int hnd_mode;
+	kern_handle_t hnd_value;
+	kern_status_t hnd_result;
+} kern_msg_handle_t;
+
+typedef struct {
+	/* transaction id. identifies a particular request/response exchange.
+	 * used when replying to a particular message. */
+	msgid_t msg_id;
+	/* the id of the task that sent a particular message. */
+	tid_t msg_sender;
+	/* the id of the port or channel used to send a particular message. */
+	koid_t msg_endpoint;
+	/* a list of iovecs that point to the buffers that make up the main
+	 * message data. */
+	kern_iovec_t *msg_data;
+	size_t msg_data_count;
+	/* a list of handle entries that contain the kernel handles included
+	 * in a message. */
+	kern_msg_handle_t *msg_handles;
+	size_t msg_handles_count;
+} kern_msg_t;
 
 #endif

syscall/msg.c

@@ -161,17 +161,17 @@ kern_status_t sys_port_disconnect(kern_handle_t port_handle)
 
 static bool validate_iovec(
 	struct task *task,
-	const struct iovec *iov,
+	const kern_iovec_t *iov,
 	size_t count,
 	bool rw)
 {
-	if (!validate_access_r(task, iov, count * sizeof(struct iovec))) {
+	if (!validate_access_r(task, iov, count * sizeof(*iov))) {
 		return false;
 	}
 
 	for (size_t i = 0; i < count; i++) {
 		bool ok = false;
-		const struct iovec *vec = &iov[i];
+		const kern_iovec_t *vec = &iov[i];
 		if (rw) {
 			ok = validate_access_w(task, vec->io_base, vec->io_len);
 		} else {
@@ -186,23 +186,55 @@ static bool validate_iovec(
 	return true;
 }
 
-kern_status_t sys_msg_send(
-	kern_handle_t port_handle,
-	const struct iovec *req_data,
-	size_t req_data_count,
-	struct iovec *resp_data,
-	size_t resp_data_count)
+static bool validate_msg(struct task *task, const kern_msg_t *msg, bool rw)
 {
-	struct task *self = current_task();
-
-	if (!validate_iovec(self, req_data, req_data_count, false)) {
+	if (!msg) {
 		return false;
 	}
 
-	if (!validate_iovec(self, resp_data, resp_data_count, true)) {
+	vm_prot_t flags;
+	if (rw) {
+		flags = VM_PROT_WRITE | VM_PROT_USER;
+	} else {
+		flags = VM_PROT_READ | VM_PROT_USER;
+	}
+
+	if (!validate_access(task, msg, sizeof *msg, flags)) {
 		return false;
 	}
 
+	if (!validate_iovec(task, msg->msg_data, msg->msg_data_count, rw)) {
+		return false;
+	}
+
+	size_t handle_buffer_len
+		= msg->msg_handles_count * sizeof(*msg->msg_handles);
+	if (!validate_access(
+		    task,
+		    msg->msg_handles,
+		    handle_buffer_len,
+		    flags)) {
+		return false;
+	}
+
+	return true;
+}
+
+kern_status_t sys_msg_send(
+	kern_handle_t port_handle,
+	const kern_msg_t *msg,
+	kern_msg_t *out_reply)
+{
+	struct task *self = current_task();
+
+	if (!validate_msg(self, msg, false)) {
+		return KERN_MEMORY_FAULT;
+	}
+
+	if (!validate_msg(self, out_reply, true)) {
+		return KERN_MEMORY_FAULT;
+	}
+
 	unsigned long flags;
 
 	task_lock_irqsave(self, &flags);
@@ -230,32 +262,18 @@ kern_status_t sys_msg_send(
 	}
 
 	port_lock_irqsave(port, &flags);
-	status = port_send_msg(
-		port,
-		req_data,
-		req_data_count,
-		resp_data,
-		resp_data_count,
-		&flags);
+	status = port_send_msg(port, msg, out_reply, &flags);
 	port_unlock_irqrestore(port, flags);
 	object_unref(port_obj);
 
 	return status;
 }
 
-kern_status_t sys_msg_recv(
-	kern_handle_t channel_handle,
-	msgid_t *out_id,
-	struct iovec *out_data,
-	size_t out_data_count)
+kern_status_t sys_msg_recv(kern_handle_t channel_handle, kern_msg_t *out_msg)
 {
 	struct task *self = current_task();
 
-	if (!validate_access_w(self, out_id, sizeof *out_id)) {
-		return KERN_MEMORY_FAULT;
-	}
-
-	if (!validate_iovec(self, out_data, out_data_count, true)) {
+	if (!validate_msg(self, out_msg, true)) {
 		return KERN_MEMORY_FAULT;
 	}
 
@@ -286,12 +304,7 @@ kern_status_t sys_msg_recv(
 	}
 
 	channel_lock_irqsave(channel, &flags);
-	status = channel_recv_msg(
-		channel,
-		out_id,
-		out_data,
-		out_data_count,
-		&flags);
+	status = channel_recv_msg(channel, out_msg, &flags);
 	channel_unlock_irqrestore(channel, flags);
 	object_unref(channel_obj);
 
@@ -301,12 +314,11 @@ kern_status_t sys_msg_recv(
 kern_status_t sys_msg_reply(
 	kern_handle_t channel_handle,
 	msgid_t id,
-	const struct iovec *reply_data,
-	size_t reply_data_count)
+	const kern_msg_t *reply)
 {
 	struct task *self = current_task();
 
-	if (!validate_iovec(self, reply_data, reply_data_count, false)) {
+	if (!validate_msg(self, reply, true)) {
 		return KERN_MEMORY_FAULT;
 	}
 
@@ -337,12 +349,7 @@ kern_status_t sys_msg_reply(
 	}
 
 	channel_lock_irqsave(channel, &flags);
-	status = channel_reply_msg(
-		channel,
-		id,
-		reply_data,
-		reply_data_count,
-		&flags);
+	status = channel_reply_msg(channel, id, reply, &flags);
 	channel_unlock_irqrestore(channel, flags);
 	object_unref(channel_obj);
 
@@ -353,7 +360,7 @@ kern_status_t sys_msg_read(
 	kern_handle_t channel_handle,
 	msgid_t id,
 	size_t offset,
-	const struct iovec *iov,
+	const kern_iovec_t *iov,
 	size_t iov_count,
 	size_t *nr_read)
 {
@@ -412,7 +419,7 @@ kern_status_t sys_msg_write(
 	kern_handle_t channel_handle,
 	msgid_t id,
 	size_t offset,
-	const struct iovec *iov,
+	const kern_iovec_t *iov,
 	size_t iov_count,
 	size_t *nr_written)
 {

vm/vm-region.c

@@ -1713,6 +1713,52 @@ kern_status_t vm_region_read_kernel(
 	return status;
 }
 
+kern_status_t vm_region_write_kernel(
+	struct vm_region *dst_region,
+	virt_addr_t dst_ptr,
+	size_t count,
+	const void *srcp,
+	size_t *nr_written)
+{
+	if (dst_region->vr_status != VM_REGION_ONLINE) {
+		return KERN_BAD_STATE;
+	}
+
+	struct vm_iterator dst;
+	const char *src = srcp;
+
+	vm_iterator_begin(
+		&dst,
+		dst_region,
+		dst_ptr,
+		VM_PROT_WRITE | VM_PROT_USER);
+
+	kern_status_t status = KERN_OK;
+	size_t r = 0;
+
+	while (r < count && dst.it_max) {
+		size_t remaining = count - r;
+		size_t to_move = MIN(dst.it_max, remaining);
+		memmove(dst.it_buf, src, to_move);
+
+		status = vm_iterator_seek(&dst, to_move);
+		if (status != KERN_OK) {
+			break;
+		}
+
+		r += to_move;
+		src += to_move;
+	}
+
+	vm_iterator_finish(&dst);
+
+	if (nr_written) {
+		*nr_written = r;
+	}
+
+	return status;
+}
+
 kern_status_t vm_region_memmove(
 	struct vm_region *dest_region,
 	virt_addr_t dest_ptr,
@@ -1775,11 +1821,11 @@ kern_status_t vm_region_memmove(
 extern kern_status_t vm_region_memmove_v(
 	struct vm_region *dest_region,
 	size_t dest_offset,
-	const struct iovec *dest_vecs,
+	const kern_iovec_t *dest_vecs,
 	size_t nr_dest_vecs,
 	struct vm_region *src_region,
 	size_t src_offset,
-	const struct iovec *src_vecs,
+	const kern_iovec_t *src_vecs,
 	size_t nr_src_vecs,
 	size_t bytes_to_move,
 	size_t *nr_bytes_moved)